Most plants never send technicians' repair notes back to their predictive maintenance models
IoT Analytics expects the smart maintenance market to grow from $8.1 billion in 2025 to $20.4 billion by 2032. The weakest link sits after the repair. In most plants, the technician's closeout note never reaches the predictive model, so work-order discipline in the CMMS matters as much as sensors or AI.
This story was produced through MarketScale. See how Industrial IoT teams put it to work with AI Visibility (GEO).
Key takeaways
The cheapest upgrade in a $20.4 billion market may be a required field on a work order that asks one question: was the fault real?
Nick Haase of MaintainX estimates that within about five years, half of maintenance and reliability professionals will have hit retirement age—making it urgent to turn closeout knowledge into structured data.
Gousto's jump from 70% to 98.5% mechanical availability, with repair time cut from one hour to 15 minutes, is a useful benchmark for what wiring alerts straight into work orders can return.
Free workspace
Turn your Industrial IoT expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
Some of the most useful data in a predictive maintenance program gets typed by a technician at the end of a repair. It says whether the fault the sensor flagged was really there. In most plants, that note never makes it back to the model that raised the alert.
That finding anchors a November 18 analysis by Zeynep Kaman of IoT Analytics, drawn from the firm's new 376-page Smart Maintenance & Asset Performance Market Report 2026, 2032. IoT Analytics puts the smart maintenance market at $8.1 billion in 2025 and projects it will reach $20.4 billion by 2032. That is a 14% CAGR, and it would make the market 2.5 times larger. Closing the loop from sensor alert to finished repair is now one of the top themes when people discuss smart maintenance.
The research has moved in the same direction. Compared with the 2023 edition, IoT Analytics widened the report's scope from predictive maintenance to the full maintenance workflow, from capturing asset-health data through diagnosing problems to planning and carrying out the maintenance response. A report that once stopped at the prediction now follows the work order.
For a maintenance and reliability leader choosing what to buy next, the most useful part of the report is where it finds the break. The step most often missing is the learning step. That weak point comes after the wrench is put down.
Six steps from alert to lesson
IoT Analytics describes the loop as a chain: a sensor spots a fault, software diagnoses it, a work order is completed, and the outcome improves the next alert. In its definition, learning sits inside the loop rather than after it. By that account, closing the loop links each stage, from the initial sensor alert through the finished repair, and then sends the outcome back into the predictive model.
For a finer-grained view, 105.kz, which writes for industrial operators in Kazakhstan, split that chain into six labeled steps in a Russian-language summary of the analysis. The labels below are 105.kz's paraphrase, translated here, not a list quoted from IoT Analytics' own text:
- Detection
- Diagnosis and prioritization
- Work order creation and scheduling
- Execution
- Closeout with documentation
- Validation and model learning
Step six is the one that goes missing most often, according to IoT Analytics. A technician's record that a defect was confirmed, or that the alarm was false, is the label a predictive model needs to get better. When that label stays in the maintenance system, or on paper, the model keeps making the same calls on the same evidence.
105.kz describes what the break looks like on the ground at oil and gas sites such as compressor and pumping stations. A vibration or temperature signal becomes an email or a phone call to a dispatcher, while the repair request lives separately in a paper or electronic log. Nothing in that path carries the outcome back to the analytics.
The gaps start earlier than step six, too. For power plants, the outlet notes, prioritization in step two needs data on asset criticality, planned outages and staff availability, not sensor readings alone. A loop is only as good as the context each step can see.
Maintenance teams never made this a priority, and for years they didn't need to. Experienced technicians covered the gap. When an alert came in, they recognized the fault, repaired it and carried the pattern forward to the next time it showed up. The learning step lived in someone's head.
That head is heading for the door. In a 2025 IoT Analytics survey of 120 digital decision-makers in machine building, 43% named workforce skill gaps and a shortage of expertise as a critical or significant barrier to adopting AI. Nick Haase, co-founder of MaintainX, estimates that within about 5 years, half of all maintenance and reliability professionals will have hit retirement age. The report's conclusion follows from that: once those experts are gone, either systems close the loop or repair times rise.
Haase put it plainly in June 2025: "They're leaving the workforce all that tribal knowledge is leaving." That tribal knowledge is, in effect, the step-six label stored in a person rather than a system. That suggests the closeout field is where a plant can start capturing it before it retires.
What a connected loop did at Gousto
Gousto linked its factory equipment to one data platform, including older machines it retrofitted with vibration sensors. Anomalies now become work orders in its CMMS, and engineers receive them through an app. In the 18 months that followed, mechanical availability climbed from 70% to 98.5%, while mean time to repair dropped from one hour to 15 minutes.
Those two metrics matter because of how downtime savings show up, or don't. IoT Analytics estimates unplanned downtime costs industry between $0.5 trillion and $1.0 trillion a year, and about 87% of that is indirect: lost revenue and missed orders rather than repairs. A failure that never happened leaves no line on the P&L, so buyers look for proof on the shop floor.
A failure that does happen gets measured in detail. LyondellBasell's propylene oxide/TBA asset in Bayport, Texas, went down in an unplanned outage in March 2026 that ran for about 13 weeks. By management's estimate, each week cost $25 million in lost EBITDA, or about $150,000 per hour, adding up to about $40 million in the first quarter and $250 million in the second.
The asymmetry is stark. An outage gets costed to the hour, while a prevented one has to be shown through operating metrics. That is why Gousto's availability and repair-time figures do the persuading.
One case is not a forecast. The 105.kz summary makes the same point: Gousto's result is a reference point, and repeating it depends on the state of a plant's data, its processes and how carefully work orders get closed. Still, the mechanism is worth noticing. IoT Analytics charts the move toward a closed loop as one break closing at a time, and Gousto's gain came from closing the early ones: connecting the anomaly to the work order and the engineer, which is steps one through four. Add disciplined closeout and validation and the model has something to learn from.
The cheapest upgrade in a $20.4 billion market may be a required field on a work order that asks one question: was the fault real?
The cheapest upgrade in a $20.4 billion market may be a required field on a work order that asks one question: was the fault real?
Vendors are buying their way across the gaps
Closing the loop end to end usually takes products from several companies, IoT Analytics found. Sensor specialists, analytics vendors and CMMS providers each cover part of it, and they are linking up through integrations, partnerships and acquisitions. The analysis profiles four vendors closing it from different starting points: Siemens, Schneider Electric, Rockwell Automation and MaintainX. The research behind that conclusion includes more than 75 expert interviews, two surveys of 120 end users each, and visits to four trade fairs, including Hannover Messe and IMTS 2026.
A crowded, fast-growing field
IoT Analytics, Smart Maintenance & Asset Performance Market Report 2026, 2032
Baker Hughes, IBM and Emerson lead the market, but each takes a different route. Baker Hughes and Emerson build on sensors they make themselves, while IBM and SAP sell only software. Wireless sensing is the fastest-growing segment and should become the largest by 2032, with specialists such as I-care, AssetWatch and Tractian gaining ground.
Wireless matters for the learning step because of where it puts sensors. IoT Analytics says wireless sensors extend continuous monitoring from critical machines to the wider asset base, and several automation incumbents are losing share in the segment. 105.kz puts the wireless segment at $1.1 billion in 2025, growing to $5.1 billion by 2032 at a 24% CAGR. More monitored machines would mean more alerts, and every one of them needs a closeout verdict before a model can learn from it.
IoT Analytics sizes six segments: three asset monitoring solutions (wireless, wired and portable vibration sensing, including bundled software and services) and three software categories (standalone predictive and prescriptive analytics, asset performance management, and CMMS). Managed monitoring, reliability consulting, EAM, ERP and MES sit outside the core scope of the $20.4 billion forecast.
Strategic buyers are consolidating the market. IoT Analytics counts 13 smart maintenance vendors acquired so far in 2026, a record. The deals include Autodesk buying MaintainX for $3.6 billion and Schneider Electric agreeing to purchase Cognite for $3.1 billion. IBM, SAP and Siemens have rolled out AI agents and copilots that convert alerts into work orders that arrive pre-filled. Still, no vendor yet runs maintenance without a human reviewing the work.
IoT Analytics interprets this as value shifting toward the steps between alert and repair, which puts dashboard-only vendors at risk of losing accounts to rivals that tie into work execution. For vendors, that interpretation holds up. For plants, it has a clear limit. One owner can combine the software, but no purchase delivers clean closeout data. According to the report, a closed loop relies on clean data and working processes more than on having the best algorithm. The 105.kz summary adds that when asset records are incomplete or work orders are closed just to tick a box, an agent will draft the wrong work orders and the model will train on the wrong labels.
Bad labels have a human cost as well. 105.kz notes that a false alarm can erode the trust of the operating staff. With agents drafting work orders that a person still reviews, that trust is what keeps the reviewer engaged rather than clicking through.
Questions for the next CMMS review
For plants already running vibration monitoring, the practical question is whether the closeout record is structured enough to be used. A free-text closeout note doesn't help a model unless someone, or something, turns it into confirmed fault, false alarm or component replaced. That conversion is a process and data design choice, and it belongs in the specification before the next contract gets signed.
Ask every vendor in the stack the same thing: when a technician closes a work order and marks the alert as a confirmed fault or a false alarm, which system receives that result, and how fast does it reach the model that raised the alert?
Two more checks follow from the report. First, if the sensors, analytics and CMMS come from different suppliers, write down who owns the integration that carries closeout data back upstream, especially if one of those suppliers changes hands. Second, measure availability and mean time to repair before any rollout, because those are the before-and-after numbers that made Gousto's case readable. The report treats asset availability and OEE as the main KPIs smart maintenance has to improve.
105.kz sketches how to test the loop before committing to it: a pilot of 8 to 12 weeks on one or two critical assets, in five steps.
- Audit the data: which signals already exist (vibration, temperature, current, SCADA, work order history, the equipment register), whether assets share one numbering scheme, and whether work orders have a field for failure cause.
- Pick a pump, compressor, mill or fan with high criticality and a clear failure history.
- Route anomalies into the CMMS, with the closeout fields described above made mandatory.
- Run the pilot for 8 to 12 weeks, collecting alarms, work orders, repair times and the share of false alarms, then compare against a pre-pilot baseline.
- Calculate the change in availability, mean time to repair and the number of prevented failures, and only then decide whether to scale.
Step three is the learning step in miniature. The false-alarm share in step four only exists if technicians are required to record it, so the pilot tests the closeout discipline as much as the sensors.
For operators, the thing to watch is whether vendors that sell both work execution software and analytics build a native path that sends confirmed-fault data back into their models. If they do, the learning step could stop being a custom integration and become a product feature. Until then, the plant that writes the requirement into its own work orders is the one whose models will actually get smarter.
Sources
Featured companies
Your experts belong here
Every story in MarketScale Industrial IoT starts with a company putting its controls engineers, plant-floor specialists, and integration partners on the record. Buyers are already reading this topic. The only question is whose experts they find.
Plant and controls buyers research deep before contact, and your engineers get to shape that research.
About the author
The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.